reverse parsing: Clearer algorithm

We used to have a kind of open-coded coroutine arrangement for parsing
reverse domains.  This is complicated and confusing.

Instead, when "parsing" the labels, simply stash their locations and
lengths.  Then, at the end, we have all of the information needed to
check, parse and process the whole domain.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>